diff --git a/requirements/common.txt b/requirements/common.txt
index 8b37f3cd30..d6e2031f53 100644
--- a/requirements/common.txt
+++ b/requirements/common.txt
@@ -7,7 +7,7 @@ requests >= 2.26.0
 tqdm
 blake3
 py-cpuinfo
-transformers >= 4.56.0, != 5.0.*, != 5.1.*, != 5.2.*, != 5.3.*, != 5.4.*, != 5.5.0
+transformers >= 5.5.3
 tokenizers >= 0.21.1  # Required for fast incremental detokenization.
 safetensors >= 0.6.2  # MXFP4/MXFP6 dtype support (F8_E8M0, F4) added in 0.6.0: https://github.com/huggingface/safetensors/pull/611
 protobuf >= 5.29.6, !=6.30.*, !=6.31.*, !=6.32.*, !=6.33.0.*, !=6.33.1.*, !=6.33.2.*, !=6.33.3.*, !=6.33.4.* # Required by LlamaTokenizer, gRPC. CVE-2026-0994
diff --git a/vllm/config/vllm.py b/vllm/config/vllm.py
index 9be5638132..a1a3420945 100644
--- a/vllm/config/vllm.py
+++ b/vllm/config/vllm.py
@@ -14,12 +14,10 @@ from dataclasses import is_dataclass
 from datetime import datetime
 from enum import IntEnum
 from functools import lru_cache
-from importlib.metadata import version
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, Literal, TypeVar, get_args
 
 import torch
-from packaging.version import Version
 from pydantic import ConfigDict, Field, model_validator
 
 import vllm.envs as envs
@@ -697,10 +695,8 @@ class VllmConfig:
         # Therefore, the presence of tie_word_embeddings in SomeVLTextConfig cannot
         # be used as a signal for whether tie_word_embeddings should be copied from
         # hf_config to the language_model config.
-        if (
-            Version(version("transformers")) >= Version("5.0.0")
-            and model_config.is_multimodal_model
-            and hasattr(model_config.hf_config, "tie_word_embeddings")
+        if model_config.is_multimodal_model and hasattr(
+            model_config.hf_config, "tie_word_embeddings"
         ):
             tie_word_embeddings = model_config.hf_config.tie_word_embeddings
             hf_config.get_text_config().tie_word_embeddings = tie_word_embeddings
diff --git a/vllm/model_executor/model_loader/weight_utils.py b/vllm/model_executor/model_loader/weight_utils.py
index dd96e15261..4ffd6b92d6 100644
--- a/vllm/model_executor/model_loader/weight_utils.py
+++ b/vllm/model_executor/model_loader/weight_utils.py
@@ -77,30 +77,13 @@ logger = init_logger(__name__)
 temp_dir = tempfile.gettempdir()
 
 
-def enable_hf_transfer():
-    """automatically activates hf_transfer"""
-    if "HF_HUB_ENABLE_HF_TRANSFER" not in os.environ:
-        try:
-            # enable hf hub transfer if available
-            import hf_transfer  # type: ignore # noqa
-
-            huggingface_hub.constants.HF_HUB_ENABLE_HF_TRANSFER = True
-        except ImportError:
-            pass
-
-
 def enable_xet_high_performance():
     """automatically activates xet high performance mode"""
     if "HF_XET_HIGH_PERFORMANCE" not in os.environ:
         huggingface_hub.constants.HF_XET_HIGH_PERFORMANCE = True
 
 
-if hasattr(huggingface_hub.constants, "HF_XET_HIGH_PERFORMANCE"):
-    # Transformers v5
-    enable_xet_high_performance()
-else:
-    # Transformers v4
-    enable_hf_transfer()
+enable_xet_high_performance()
 
 
 class DisabledTqdm(tqdm):
diff --git a/vllm/model_executor/models/gemma3n_mm.py b/vllm/model_executor/models/gemma3n_mm.py
index 2b5266f0c9..1dd44313c1 100644
--- a/vllm/model_executor/models/gemma3n_mm.py
+++ b/vllm/model_executor/models/gemma3n_mm.py
@@ -618,13 +618,8 @@ class Gemma3nForConditionalGeneration(
         input_features = audio_input["input_features_padded"].squeeze(1)
         input_features_mask = audio_input["input_features_mask"].squeeze(1)
         audio_outputs = self.audio_tower(input_features, ~input_features_mask)
-        if isinstance(audio_outputs, tuple):
-            # Transformers v4
-            audio_encodings, audio_mask = audio_outputs
-        else:
-            # Transformers v5
-            audio_encodings = audio_outputs.last_hidden_state
-            audio_mask = audio_outputs.audio_mel_mask
+        audio_encodings = audio_outputs.last_hidden_state
+        audio_mask = audio_outputs.audio_mel_mask
         audio_features = self.embed_audio(inputs_embeds=audio_encodings)
 
         # The Gemma3nProcessor expects all audio will be 30s in length and
diff --git a/vllm/model_executor/models/qwen3_omni_moe_thinker.py b/vllm/model_executor/models/qwen3_omni_moe_thinker.py
index 05586324df..f37ecc0ed2 100755
--- a/vllm/model_executor/models/qwen3_omni_moe_thinker.py
+++ b/vllm/model_executor/models/qwen3_omni_moe_thinker.py
@@ -30,9 +30,7 @@ import numpy as np
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-from packaging.version import Version
 from transformers import PretrainedConfig
-from transformers import __version__ as TRANSFORMERS_VERSION
 from transformers.feature_extraction_utils import BatchFeature
 from transformers.models.qwen3_omni_moe.configuration_qwen3_omni_moe import (
     Qwen3OmniMoeAudioEncoderConfig,
@@ -1261,40 +1259,6 @@ class Qwen3OmniMoeThinkerMultiModalProcessor(
             tok_kwargs = dict(tok_kwargs)
             mm_kwargs["audio_kwargs"] = dict(mm_kwargs.get("audio_kwargs") or {})
             mm_kwargs["text_kwargs"] = dict(mm_kwargs.get("text_kwargs") or {})
-            if Version(TRANSFORMERS_VERSION) < Version("4.58.0"):
-                # Extract audio_sample_rate before restructuring
-                audio_sample_rate = mm_kwargs.pop("audio_sample_rate", None)
-
-                # move truncation to audio_kwargs level to avoid conflict
-                # with tok_kwargs
-                mm_kwargs["audio_kwargs"].setdefault(
-                    "truncation", mm_kwargs.pop("truncation", False)
-                )
-                mm_kwargs["text_kwargs"].setdefault(
-                    "truncation", tok_kwargs.pop("truncation", False)
-                )
-
-                # Validate and conditionally pass audio_sample_rate
-                # WhisperFeatureExtractor has a fixed sampling rate, and vLLM's
-                # audio loader already resamples audio to the target rate.
-                # Only pass the value if it matches to avoid unexpected behavior.
-                if audio_sample_rate is not None:
-                    expected_sr = feature_extractor.sampling_rate
-                    if audio_sample_rate != expected_sr:
-                        logger.warning(
-                            "[%s] audio_sample_rate mismatch: user provided %dHz "
-                            "but model expects %dHz. Ignoring user value. "
-                            "vLLM's audio loader already resampled to %dHz.",
-                            self.__class__.__name__,
-                            audio_sample_rate,
-                            expected_sr,
-                            expected_sr,
-                        )
-                    else:
-                        # Sample rate matches, safe to pass
-                        mm_kwargs["audio_kwargs"]["audio_sample_rate"] = (
-                            audio_sample_rate
-                        )
 
         hf_inputs = super()._call_hf_processor(
             prompt=prompt,
diff --git a/vllm/model_executor/models/transformers/base.py b/vllm/model_executor/models/transformers/base.py
index 35897ce7db..234ae9570b 100644
--- a/vllm/model_executor/models/transformers/base.py
+++ b/vllm/model_executor/models/transformers/base.py
@@ -27,6 +27,10 @@ import transformers
 from packaging.version import Version
 from torch import nn
 from transformers import AutoModel
+from transformers.conversion_mapping import (
+    WeightRenaming,
+    get_model_conversion_mapping,
+)
 from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS
 
 from vllm.compilation.decorators import support_torch_compile
@@ -212,16 +216,9 @@ class Base(
         `create_attention_instances` are used
         - Sets the dtype to the default torch dtype set by vLLM because Transformers
         uses the config dtype when creating the model
-        - Propagates this dtype to any sub-configs because Transformers model
-        implementations do not support/use different dtypes in sub-models
         """
         self.text_config._attn_implementation = "vllm"
         self.config.dtype = torch.get_default_dtype()
-        # TODO(hmellor): Remove this when Transformers v4 support is dropped
-        for sub_config_name in getattr(self.config, "sub_configs", {}):
-            sub_config = getattr(self.config, sub_config_name)
-            if sub_config.dtype != (dtype := self.config.dtype):
-                sub_config.dtype = dtype
 
     def _get_decoder_cls(self, **kwargs: dict) -> type[PreTrainedModel]:
         """
@@ -300,9 +297,7 @@ class Base(
 
         This handles:
 
-        - Transformers weight renaming:
-            - from `WeightRenaming` in Transformers v5
-            - from `_checkpoint_conversion_mapping` in Transformers v4
+        - Transformers weight renaming from `WeightRenaming`
         - Checkpoints saved with a base model prefix that is not `model`
         - Checkpoints saved with no base model prefix
         - Any quantization config specific mappings
@@ -310,37 +305,16 @@ class Base(
         self.hf_to_vllm_mapper = WeightsMapper()
         orig_to_new_regex = self.hf_to_vllm_mapper.orig_to_new_regex
 
-        if Version(transformers.__version__) >= Version("5.0.0"):
-            from transformers.conversion_mapping import (
-                WeightRenaming,
-                get_model_conversion_mapping,
-            )
-
-            for mapping in get_model_conversion_mapping(self.model):
-                # Handle weights which have been renamed in Transformers
-                if isinstance(mapping, WeightRenaming):
-                    # Recompile using regex (Transformers used re)
-                    compiled_sources = re.compile(
-                        mapping.compiled_sources.pattern, mapping.compiled_sources.flags
-                    )
-                    target_pattern = mapping.target_patterns[0]
-                    orig_to_new_regex[compiled_sources] = target_pattern
-                # TODO: Handle WeightConverter to enable layer merging
-        else:
-            # Replace legacy suffixes used for norms
-            # TODO(hmellor): Remove this when Transformers v4 support is dropped
-            orig_to_new_regex.update(
-                {
-                    re.compile(r"\.gamma$"): ".weight",
-                    re.compile(r"\.beta$"): ".bias",
-                }
-            )
-
-        # Handle weights which have been renamed in Transformers
-        # TODO(hmellor): Remove this when Transformers v4 support is dropped
-        ccm = getattr(self.model, "_checkpoint_conversion_mapping", {})
-        for source, target in ccm.items():
-            orig_to_new_regex[re.compile(source)] = target
+        for mapping in get_model_conversion_mapping(self.model):
+            # Handle weights which have been renamed in Transformers
+            if isinstance(mapping, WeightRenaming):
+                # Recompile using regex (Transformers used re)
+                compiled_sources = re.compile(
+                    mapping.compiled_sources.pattern, mapping.compiled_sources.flags
+                )
+                target_pattern = mapping.target_patterns[0]
+                orig_to_new_regex[compiled_sources] = target_pattern
+            # TODO: Handle WeightConverter to enable layer merging
 
         # Handle unexpected weights which should be ignored
         if self.model._keys_to_ignore_on_load_unexpected is not None:
@@ -377,7 +351,7 @@ class Base(
         """
         Check if the model has tied word embeddings.
         """
-        # Transformers v4 and v5 will store this in different places
+        # Models created with Transformers v4 and v5 will store this in different places
         tie_word_embeddings_v4 = getattr(self.text_config, "tie_word_embeddings", False)
         tie_word_embeddings_v5 = getattr(self.config, "tie_word_embeddings", False)
         return tie_word_embeddings_v4 or tie_word_embeddings_v5
diff --git a/vllm/model_executor/models/transformers/utils.py b/vllm/model_executor/models/transformers/utils.py
index dbf0a084f7..0a4ca94c5e 100644
--- a/vllm/model_executor/models/transformers/utils.py
+++ b/vllm/model_executor/models/transformers/utils.py
@@ -101,8 +101,6 @@ Style = Literal[
     "replicate",
     "colwise_gather_output",
     "rowwise_split_input",
-    "colwise_rep",
-    "rowwise_rep",
 ]
 
 
@@ -131,12 +129,8 @@ def replace_linear_class(
         "colwise": (ColumnParallelLinear, {}),
         "rowwise": (RowParallelLinear, {}),
         "replicate": (ReplicatedLinear, {}),
-        # Transformers v5
         "colwise_gather_output": (ColumnParallelLinear, {"gather_output": True}),
         "rowwise_split_input": (RowParallelLinear, {"input_is_parallel": False}),
-        # Transformers v4
-        "colwise_rep": (ColumnParallelLinear, {"gather_output": True}),
-        "rowwise_rep": (RowParallelLinear, {"input_is_parallel": False}),
     }.get(style, (ReplicatedLinear, {}))
 
     return vllm_linear_cls(
diff --git a/vllm/model_executor/models/ultravox.py b/vllm/model_executor/models/ultravox.py
index 986255d86f..2c1b2b02e8 100644
--- a/vllm/model_executor/models/ultravox.py
+++ b/vllm/model_executor/models/ultravox.py
@@ -5,7 +5,6 @@
 """PyTorch Ultravox model."""
 
 import copy
-import inspect
 from collections.abc import Iterable, Mapping, Sequence
 from types import SimpleNamespace
 from typing import Annotated, Any, Literal, TypeAlias
@@ -397,17 +396,10 @@ class UltravoxTransformerProjector(nn.Module, ModuleUtilsMixin):
         )
         hidden_states = hidden_states + positions
 
-        # Backward compatibility for Transformers v4 where layer_head_mask
-        # was a required argument for WhisperEncoderLayer.forward
-        kwargs = {}
-        if "layer_head_mask" in inspect.signature(self.layers[0].forward).parameters:
-            kwargs["layer_head_mask"] = None
-
         for layer in self.layers:
             hidden_states = layer(
                 hidden_states,
                 attention_mask=extended_attention_mask,
-                **kwargs,
             )
             # BC version that allows for the old tupled output
             if isinstance(hidden_states, tuple):
@@ -504,17 +496,10 @@ class ModifiedWhisperEncoder(WhisperEncoder):
 
         attention_mask = self.get_attention_mask_by_audio_len(audio_lens, hidden_states)
 
-        # Backward compatibility for Transformers v4 where layer_head_mask
-        # was a required argument for WhisperEncoderLayer.forward
-        kwargs = {}
-        if "layer_head_mask" in inspect.signature(self.layers[0].forward).parameters:
-            kwargs["layer_head_mask"] = None
-
         for encoder_layer in self.layers:
             hidden_states = encoder_layer(
                 hidden_states,
                 attention_mask,
-                **kwargs,
             )
             # BC version that allows for the old tupled output
             if isinstance(hidden_states, tuple):
diff --git a/vllm/tokenizers/mistral.py b/vllm/tokenizers/mistral.py
index 8fce690433..8e29e1e5d6 100644
--- a/vllm/tokenizers/mistral.py
+++ b/vllm/tokenizers/mistral.py
@@ -31,21 +31,13 @@ from mistral_common.tokens.tokenizers.sentencepiece import (
 )
 from mistral_common.tokens.tokenizers.tekken import Tekkenizer
 from pydantic import ValidationError
+from transformers.tokenization_mistral_common import MistralCommonBackend
 
 from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
 from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest
 from vllm.logger import init_logger
 from vllm.tokenizers.protocol import TokenizerLike
 
-try:
-    # Transformers v5
-    from transformers.tokenization_mistral_common import MistralCommonBackend
-except ImportError:
-    # Transformers v4
-    from transformers.tokenization_mistral_common import (
-        MistralCommonTokenizer as MistralCommonBackend,
-    )
-
 if TYPE_CHECKING:
     import llguidance
     from transformers import BatchEncoding
diff --git a/vllm/transformers_utils/config.py b/vllm/transformers_utils/config.py
index a5878ca028..427f30b399 100644
--- a/vllm/transformers_utils/config.py
+++ b/vllm/transformers_utils/config.py
@@ -16,6 +16,7 @@ from huggingface_hub import constants
 from packaging.version import Version
 from safetensors.torch import _TYPES as _SAFETENSORS_TO_TORCH_DTYPE
 from transformers import GenerationConfig, PretrainedConfig
+from transformers.configuration_utils import ALLOWED_LAYER_TYPES
 from transformers.models.auto.image_processing_auto import get_image_processor_config
 from transformers.models.auto.modeling_auto import (
     MODEL_FOR_CAUSAL_LM_MAPPING_NAMES,
@@ -49,15 +50,6 @@ from .repo_utils import (
     with_retry,
 )
 
-try:
-    # Transformers v5
-    from transformers.configuration_utils import ALLOWED_ATTENTION_LAYER_TYPES
-except ImportError:
-    # Transformers v4
-    from transformers.configuration_utils import (
-        ALLOWED_LAYER_TYPES as ALLOWED_ATTENTION_LAYER_TYPES,
-    )
-
 if envs.VLLM_USE_MODELSCOPE:
     from modelscope import AutoConfig
 else:
@@ -68,9 +60,8 @@ MISTRAL_CONFIG_NAME = "params.json"
 logger = init_logger(__name__)
 
 if Version(version("transformers")) < Version("5.0.0"):
-    logger.warning(
-        "Support for Transformers v4 is deprecated. The Transformers v4 codepath will "
-        "become unmaintained in vLLM v0.22.0 and will be removed in vLLM v0.24.0. "
+    raise ImportError(
+        "Support for Transformers v4 is deprecated and was removed in vLLM v0.24.0. "
         "Please upgrade to Transformers v5: pip install --upgrade transformers"
     )
 
@@ -159,7 +150,7 @@ def is_rope_parameters_nested(rope_parameters: dict[str, Any]) -> bool:
     # Cannot be nested if rope_parameters is empty
     if not rope_parameters:
         return False
-    return set(rope_parameters.keys()).issubset(ALLOWED_ATTENTION_LAYER_TYPES)
+    return set(rope_parameters.keys()).issubset(ALLOWED_LAYER_TYPES)
 
 
 @contextmanager
@@ -183,24 +174,21 @@ def _patch_hf_transformers_validate_rope():
     hf transformers (from v5 onwards)
     """
 
-    if Version(version("transformers")) >= Version("5.0.0"):
-        if hasattr(PretrainedConfig.validate_rope, "__vllm_patched__"):
-            return
+    if hasattr(PretrainedConfig.validate_rope, "__vllm_patched__"):
+        return
 
-        _original_validate_rope = PretrainedConfig.validate_rope
+    _original_validate_rope = PretrainedConfig.validate_rope
 
-        @wraps(_original_validate_rope)
-        def patched_validate_rope(self, *args, **kwargs):
-            ignore_keys_param = kwargs.pop("ignore_keys", None)
-            original_ignore_keys = self.ignore_keys_at_rope_validation
-            self.ignore_keys_at_rope_validation = (
-                original_ignore_keys or ignore_keys_param
-            )
-            result = _original_validate_rope(self, *args, **kwargs)
-            return result
+    @wraps(_original_validate_rope)
+    def patched_validate_rope(self, *args, **kwargs):
+        ignore_keys_param = kwargs.pop("ignore_keys", None)
+        original_ignore_keys = self.ignore_keys_at_rope_validation
+        self.ignore_keys_at_rope_validation = original_ignore_keys or ignore_keys_param
+        result = _original_validate_rope(self, *args, **kwargs)
+        return result
 
-        patched_validate_rope.__vllm_patched__ = True  # type: ignore[attr-defined]
-        PretrainedConfig.validate_rope = patched_validate_rope
+    patched_validate_rope.__vllm_patched__ = True  # type: ignore[attr-defined]
+    PretrainedConfig.validate_rope = patched_validate_rope
 
 
 class HFConfigParser(ConfigParserBase):
@@ -493,39 +481,13 @@ def patch_rope_parameters(config: PretrainedConfig) -> None:
     """Provide backwards compatibility for RoPE."""
     from vllm.config.utils import getattr_iter
 
-    # Older custom models may use non-standard field names
-    # which need patching for both Transformers v4 and v5.
+    # Older custom models may use non-standard field names which need patching.
     names = ["rope_theta", "rotary_emb_base"]
     rope_theta = getattr_iter(config, names, None, warn=True)
     names = ["partial_rotary_factor", "rotary_pct", "rotary_emb_fraction"]
     partial_rotary_factor = getattr_iter(config, names, None, warn=True)
-    ompe = getattr(config, "original_max_position_embeddings", None)
 
-    if Version(version("transformers")) < Version("5.0.0"):
-        # Transformers v4 installed, legacy config fields may be present.
-        if is_rope_parameters_nested(getattr(config, "rope_parameters", {})):
-            # Loading nested rope_parameters (from Transformers v5) in Transformers v4.
-            # Skip legacy patching since it should already be in the correct format.
-            pass
-        else:
-            if (rope_scaling := getattr(config, "rope_scaling", None)) is not None:
-                config.rope_parameters = rope_scaling
-            if (
-                rope_theta is not None
-                or partial_rotary_factor is not None
-                or ompe is not None
-            ) and not getattr(config, "rope_parameters", None):
-                config.rope_parameters = {"rope_type": "default"}
-            # Patch legacy fields into rope_parameters
-            if rope_theta is not None:
-                config.rope_parameters["rope_theta"] = rope_theta
-            if partial_rotary_factor is not None:
-                config.rope_parameters["partial_rotary_factor"] = partial_rotary_factor
-            if ompe is not None:
-                config.rope_parameters["original_max_position_embeddings"] = ompe
-            patch_legacy_rope_type(getattr(config, "rope_parameters", None))
-    elif rope_theta is not None or getattr(config, "rope_parameters", None):
-        # Transformers v5 installed
+    if rope_theta is not None or getattr(config, "rope_parameters", None):
         # Patch these fields in case they used non-standard names
         if rope_theta is not None:
             config.rope_theta = rope_theta
diff --git a/vllm/transformers_utils/configs/deepseek_vl2.py b/vllm/transformers_utils/configs/deepseek_vl2.py
index 3d3e20fea8..9345306aba 100644
--- a/vllm/transformers_utils/configs/deepseek_vl2.py
+++ b/vllm/transformers_utils/configs/deepseek_vl2.py
@@ -3,6 +3,7 @@
 
 # adapted from https://github.com/deepseek-ai/DeepSeek-VL2/blob/faf18023f24b962b32d9f0a2d89e402a8d383a78/deepseek_vl2/models/modeling_deepseek_vl_v2.py#L115-L268
 
+from huggingface_hub.dataclasses import strict
 from transformers import DeepseekV2Config, PretrainedConfig
 
 
@@ -87,16 +88,9 @@ class MlpProjectorConfig(PretrainedConfig):
         super().__init__(**kwargs)
 
 
-if hasattr(DeepseekV2Config, "validate"):
-    # Transformers v5
-    from huggingface_hub.dataclasses import strict
-
-    @strict
-    class DeepseekVLV2TextConfig(DeepseekV2Config):
-        kv_lora_rank: int | None = None
-else:
-    # Transformers v4
-    DeepseekVLV2TextConfig = DeepseekV2Config  # type: ignore[misc]
+@strict
+class DeepseekVLV2TextConfig(DeepseekV2Config):
+    kv_lora_rank: int | None = None
 
 
 class DeepseekVLV2Config(PretrainedConfig):
diff --git a/vllm/transformers_utils/configs/olmo_hybrid.py b/vllm/transformers_utils/configs/olmo_hybrid.py
index 2a60f29025..cdca81757e 100644
--- a/vllm/transformers_utils/configs/olmo_hybrid.py
+++ b/vllm/transformers_utils/configs/olmo_hybrid.py
@@ -228,15 +228,8 @@ class OlmoHybridConfig(PretrainedConfig):
             if "full_attention" not in layer_types:
                 layer_types[-1] = "full_attention"
 
-        if hasattr(self, "validate_layer_type"):
-            # Transformers v5
-            self.layer_types = layer_types
-            self.validate_layer_type()
-        else:
-            # Transformers v4
-            from transformers.configuration_utils import layer_type_validation
-
-            layer_type_validation(layer_types, num_hidden_layers)
+        self.layer_types = layer_types
+        self.validate_layer_type()
         if "linear_attention" not in layer_types:
             raise ValueError(
                 "OLMoHybrid expects at least one 'linear_attention' layer."
diff --git a/vllm/transformers_utils/configs/qwen3_5.py b/vllm/transformers_utils/configs/qwen3_5.py
index 3192e5e9a1..d5820a5783 100644
--- a/vllm/transformers_utils/configs/qwen3_5.py
+++ b/vllm/transformers_utils/configs/qwen3_5.py
@@ -94,18 +94,11 @@ class Qwen3_5TextConfig(PretrainedConfig):
                 else "full_attention"
                 for i in range(self.num_hidden_layers)
             ]
-        if hasattr(self, "validate_layer_type"):
-            # Transformers v5
-            kwargs["ignore_keys_at_rope_validation"] = {
-                "mrope_section",
-                "mrope_interleaved",
-            }
-            self.validate_layer_type()
-        else:
-            # Transformers v4
-            from transformers.configuration_utils import layer_type_validation
-
-            layer_type_validation(self.layer_types, self.num_hidden_layers)
+        kwargs["ignore_keys_at_rope_validation"] = {
+            "mrope_section",
+            "mrope_interleaved",
+        }
+        self.validate_layer_type()
 
         # linear attention part
         self.linear_conv_kernel_dim = linear_conv_kernel_dim
diff --git a/vllm/transformers_utils/configs/qwen3_5_moe.py b/vllm/transformers_utils/configs/qwen3_5_moe.py
index 9d9987ce03..ec229ce814 100644
--- a/vllm/transformers_utils/configs/qwen3_5_moe.py
+++ b/vllm/transformers_utils/configs/qwen3_5_moe.py
@@ -100,18 +100,11 @@ class Qwen3_5MoeTextConfig(PretrainedConfig):
                 else "full_attention"
                 for i in range(self.num_hidden_layers)
             ]
-        if hasattr(self, "validate_layer_type"):
-            # Transformers v5
-            kwargs["ignore_keys_at_rope_validation"] = {
-                "mrope_section",
-                "mrope_interleaved",
-            }
-            self.validate_layer_type()
-        else:
-            # Transformers v4
-            from transformers.configuration_utils import layer_type_validation
-
-            layer_type_validation(self.layer_types, self.num_hidden_layers)
+        kwargs["ignore_keys_at_rope_validation"] = {
+            "mrope_section",
+            "mrope_interleaved",
+        }
+        self.validate_layer_type()
 
         # linear attention part
         self.linear_conv_kernel_dim = linear_conv_kernel_dim
diff --git a/vllm/transformers_utils/configs/qwen3_next.py b/vllm/transformers_utils/configs/qwen3_next.py
index 6a02476fbe..de579ed2cf 100644
--- a/vllm/transformers_utils/configs/qwen3_next.py
+++ b/vllm/transformers_utils/configs/qwen3_next.py
@@ -252,14 +252,7 @@ class Qwen3NextConfig(PretrainedConfig):
                 "linear_attention" if bool((i + 1) % 4) else "full_attention"
                 for i in range(self.num_hidden_layers)
             ]
-        if hasattr(self, "validate_layer_type"):
-            # Transformers v5
-            self.validate_layer_type()
-        else:
-            # Transformers v4
-            from transformers.configuration_utils import layer_type_validation
-
-            layer_type_validation(self.layer_types)
+        self.validate_layer_type()
 
         # linear attention part
         self.linear_conv_kernel_dim = linear_conv_kernel_dim
diff --git a/vllm/transformers_utils/configs/speculators/base.py b/vllm/transformers_utils/configs/speculators/base.py
index f09173bcb9..08368d346f 100644
--- a/vllm/transformers_utils/configs/speculators/base.py
+++ b/vllm/transformers_utils/configs/speculators/base.py
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 import os
-from dataclasses import fields, is_dataclass
+from dataclasses import fields
 from typing import Any
 
 from transformers import PretrainedConfig
@@ -16,11 +16,8 @@ class SpeculatorsConfig(PretrainedConfig):
     model_type = "speculators"
 
     def __init__(self, **kwargs):
-        # Transformers v4 - super().__init__ which sets all kwargs as attributes
-        if not is_dataclass(PretrainedConfig):
-            return super().__init__(**kwargs)
-        # Transformers v5 - super().__init__ performs some validation before
-        # setting all kwargs as attributes, so we set them first to be safe
+        # super().__init__ performs some validation before setting all kwargs as
+        # attributes, so we set them first to be safe
         pre_trained_config_fields = {f.name for f in fields(PretrainedConfig)}
         super_kwargs = dict()
         for key, value in kwargs.items():
diff --git a/vllm/transformers_utils/processor.py b/vllm/transformers_utils/processor.py
index ec01f65d77..d0fc5c25a4 100644
--- a/vllm/transformers_utils/processor.py
+++ b/vllm/transformers_utils/processor.py
@@ -59,10 +59,6 @@ def _transformers_v4_compatibility_init() -> Any:
 
     This can be removed if `Molmo2ForConditionalGeneration` is upstreamed to
     Transformers."""
-    # Transformers v4
-    if hasattr(ProcessorMixin, "optional_attributes"):
-        return
-    # Transformers v5
     if hasattr(ProcessorMixin.__init__, "_vllm_patched"):
         return
 
diff --git a/vllm/transformers_utils/processors/pixtral.py b/vllm/transformers_utils/processors/pixtral.py
index 63c75151fc..67f0dd4b07 100644
--- a/vllm/transformers_utils/processors/pixtral.py
+++ b/vllm/transformers_utils/processors/pixtral.py
@@ -56,11 +56,6 @@ class MistralCommonPixtralProcessor(ProcessorMixin):
         image_processor: MistralCommonImageProcessor,
     ) -> None:
         self.tokenizer = tokenizer.transformers_tokenizer
-
-        # Back-compatibility for Transformers v4
-        if not hasattr(self.tokenizer, "init_kwargs"):
-            self.tokenizer.init_kwargs = {}
-
         self.image_processor = image_processor
 
         image_special_ids = self.image_processor.mm_encoder.special_ids
diff --git a/vllm/transformers_utils/processors/voxtral.py b/vllm/transformers_utils/processors/voxtral.py
index 3abe660611..f67bfe9d2e 100644
--- a/vllm/transformers_utils/processors/voxtral.py
+++ b/vllm/transformers_utils/processors/voxtral.py
@@ -111,11 +111,6 @@ class MistralCommonVoxtralProcessor(ProcessorMixin):
         feature_extractor: MistralCommonFeatureExtractor,
     ) -> None:
         self.tokenizer = tokenizer.transformers_tokenizer
-
-        # Back-compatibility for Transformers v4
-        if not hasattr(self.tokenizer, "init_kwargs"):
-            self.tokenizer.init_kwargs = {}
-
         self.feature_extractor = feature_extractor
 
         audio_special_ids = self.feature_extractor.audio_encoder.special_ids